2004 UC Proceedings Abstract
Classification of Image Textures Using GRID and a Neural Network Track: Modeling Author(s): Geoffrey Phelps Image texture can be described by statistical parameters that summarize the local neighborhood of pixels within a texture. Individual pixels in a given texture may have different local statistical parameters, but often the pattern of local statistical parameters for a given texture is unique. Samples of local statistical parameters for textures of geologic units in a DOQ are used to train an artificial neural network. For each point on the image the DOQ gray-scale values in expanding moving-window neighborhoods, as a function of distance, form the basis of an empirical function by which the network is trained. Supervised classification by the neural network then distinguishes textures in the image. The neural network also has the capability of rejecting textures that do not resemble the textures used in training, which allows for the identification of new textures. Geoffrey Phelps USGS Geophysics 345 Middlefield Road MS 989 Menlo Park , CA 94025 US Phone: 650-329-4922 E-mail: gphelps@usgs.gov |
|||||
